Insights. Performance. Applied AI.

From business questions
to better ways of working.

I connect customer understanding, measurement and practical delivery so the work ends in a useful decision and a clear next step.

The role changes the emphasis. The discipline stays the same: understand the problem, establish evidence, act, measure and learn.

Explore the first 90 days

One method, different applications

Understand the decision.
Establish the baseline.
Then choose what to change.

A dashboard should help someone act. A media plan should explain the choices and assumptions behind the budget. An AI assistant should improve the whole task. In both cases, the test is whether the work leads to a better outcome for the customer, the team or the business.

A starting framework, not a fixed promise

The first 90 days.
Shaped by the remit.

Choose the pathway closest to the role. Each has its own priorities, deliverables and decision gates. Timing depends on access, maturity and risk; the case studies illustrate the principles, not a claim that every project followed this timetable.

Pathway 01 / Insights & performance

Make the “so what”
part of the work.

For marketing analytics, customer insight and digital product improvement roles. The goal is to connect customer understanding and reliable measurement with decisions about the offer, experience and investment.

Leadership track / starts on day one

Agree who needs the insight.
And who can act on it.

Make the commercial question explicit

Align on customer and business outcomes, not a list of dashboards. Define decision rights, success measures and the stakeholders who will use the analysis.

Connect the people and the evidence

Bring research, analytics, marketing, product and delivery partners into the same priorities. Agree access, privacy boundaries, ownership and capacity to implement recommendations.

Days 1 to 30 / Understand and prioritise

Find the decision behind the request.

Meet the people responsible for marketing, product, customer experience and commercial outcomes. Ask what they need to decide, which customer problems matter and what evidence they currently trust. Combine their questions with customer research and a review of the journey.

Review measurement coverage, reporting definitions, attribution limits and access. Establish a baseline for the priority question, then rank opportunities by customer value, business value, confidence and delivery effort.

What the team has

Stakeholder questions and customer evidence map
Measurement audit, baseline and agreed success measures
Prioritised insight and improvement backlog

Ready to move on when

The team agrees which decision matters first, what evidence is missing and who can act on the answer.

Days 31 to 60 / Investigate and improve

Turn the evidence into a next step.

Investigate the priority journey using segmentation, funnel analysis and qualitative research. Look at audience needs, acquisition quality, the offer and points of hesitation. Separate a tracking problem from a behaviour change before recommending action.

Translate missing evidence into a measurement plan and technical requirements. Coordinate implementation and QA with delivery partners. Present the analysis as a decision brief: what changed, why it matters, possible explanations and what to investigate or test next.

What the team has

Focused analysis and prioritised recommendations
Measurement specifications and acceptance checks
A decision-led report and a test or improvement brief

Ready to move on when

The recommendation has an owner, a clear evidence trail and a practical way to assess whether the change helps.

Days 61 to 90 / Evaluate and embed

Make insight part of how the team works.

Review the changes with marketing and product owners. Compare suitable periods, audiences and journey stages, accounting for seasonality, channel mix and measurement changes. Use experiments where feasible; make the limits of observational comparisons explicit.

Establish a recurring decision and performance review. Record what the team changed, what it learned and what needs attention next. Prioritise the next quarter’s opportunities, including the measurement and delivery capacity they require.

What the team has

Before-and-after readout with uncertainty stated
Decision log, reporting cadence and accountable owners
Next-quarter improvement and measurement roadmap

Ready to move on when

The team can explain what it learned, what it will do next and how it will monitor the result.

What success looks like

Clearer decisions.
A repeatable improvement cycle.

Judge the work by the decisions it informs and the customer or business outcomes that follow. Use appropriate measures such as journey completion, acquisition quality, retention or service effort, with a clear baseline. Dashboard usage alone is not the result.

Adapt the plan to the remit

For a customer-insight role, put more weight on research design, audience needs and translating findings into the offer. For marketing analytics, examine channel quality, attribution and commercial effectiveness. For product improvement, emphasise journey friction, experimentation and prioritisation with product owners.

Where data quality is weak, establish what is safe to conclude before rushing to a performance claim. Where the foundation is mature, spend more of the first 90 days on diagnosis, experiments and adoption.

GOV.UK: connect performance analysis with user research

Crown Resorts

Measurement strategy

From stakeholder questions to collection, analysis and recommendations.

Explore the case

Hab Shifa

Customer-led growth

Connect customer understanding with campaign direction and team delivery.

Explore the case

News Corp

Compare Money

Translate different stakeholder needs into a shared measurement foundation.

Explore the case

Choose a pathway

Pathway 02 / Applied AI

From AI opportunity
to working capability.

Run two connected tracks: prove a useful change in the work and establish the conditions for wider adoption. Access, complexity and risk determine the pace.

Leadership track / starts on day one

Build the capability around the pilot.

Agree the mandate

Clarify the executive sponsor, business priorities, decision rights and budget. Identify where AI could improve service, growth or cost, and where a simpler change would be better.

Understand what already exists

Inventory current AI use, suppliers, data access, skills and delivery commitments, including informal use. Prioritise a small opportunity portfolio by value, feasibility, risk and readiness.

Set proportionate controls

Work with security, privacy and legal owners on approved uses, data handling, evaluation and human authority. Assign an owner to each system, with incident escalation and a way to stop or roll back.

Prepare people to own the work

Involve affected teams early. Establish training, feedback and support responsibilities, then make explicit build, buy and partner choices. Internal ownership needs time and capability, not just access to a tool.

Days 1 to 30 / Map and choose

Find a problem worth solving.

I follow a real task from request to result with the people doing the work, the people checking it and the team that will support it. Where does it wait? What information is missing? Which decisions need judgement?

Establish hands-on effort, elapsed time, corrections and handoffs. Compare opportunities against their usefulness, available information, risk and ownership before choosing a pilot.

What the team has

Workflow and ownership map
Baseline and prioritised opportunities
Pilot brief with boundaries and acceptance criteria

Ready to move on when

We can explain what we are testing, why it matters and how we will recognise improvement.

Days 31 to 60 / Build and test

Make a small version useful.

Build within the approved environment and reuse capabilities the team already has. Use rules for predictable checks, AI for retrieval, drafting and synthesis, and people for decisions that require authority.

Test ordinary work and difficult cases: incomplete input, conflicting sources and requests the system should stop or escalate. The people using it assess the result alongside the existing process.

What the team has

Working pilot and representative test cases
Evaluation results and known limitations
Review, escalation and recovery steps with a named owner

Ready to move on when

The pilot meets the agreed quality standard and can be used safely alongside the existing process.

Days 61 to 90 / Measure and decide

Decide what deserves to continue.

Compare like-for-like work. Look at completion time, reviewer effort and rework, not just generation speed. Check whether people return to it and whether cost, reliability and support are workable.

The decision may be to expand, narrow, redesign or stop. A useful pilot produces evidence for that decision, even when the answer is to change direction.

What the team has

Pilot readout: what changed and what still fails
Recommendation to expand, narrow, redesign or stop
Ownership and resourcing for the next stage

Ready to move on when

The benefit holds in real use, the risks are understood and someone owns ongoing support.

What counts as improvement

Measure the work.
Not the novelty.

Agree the baseline and acceptance criteria before the pilot. Report the benefit, review burden, adoption, operating cost and unresolved risk together. Hours released are capacity, not automatically cash savings.

Less total effort

Include briefing, checking, correction and handover. Time saved in one step can disappear in the next.

Quality people can rely on

Check accuracy, source support and whether the result helps someone make the right decision.

A process people use

Look for repeat use and workable review, rather than treating a successful demonstration as adoption.

An operating model that holds

Account for platform cost, maintenance, exceptions and the person responsible when something fails.

Building internal capability

Bring the learning closer.
Transfer ownership deliberately.

In-house development can shorten the feedback loop between the people building a tool and the people using it. It also creates obligations: documentation, maintenance, access control and support.

I work across discovery, workflow design, assistant configuration, practical automation, evaluation and adoption. Security, complex infrastructure, legal decisions and production assurance need the relevant specialists.

Own what creates a lasting advantage. Use established products where they fit. Keep specialist partners where their depth matters.

The day-90 leadership decision

A funded next step.
An owner for the outcome.

The readout should recommend what to scale, what to stop and what needs more evidence. It should include a prioritised six-to-twelve-month roadmap, expected benefits, dependencies, delivery capacity and a budget for ongoing operation.

Agree who owns each benefit, who accepts residual risk and when progress will be reviewed. Monitoring, evaluation and staff feedback continue after launch, including when models, suppliers or the underlying process change.

Adapt the plan to the role

For an organisation starting out, emphasise safe use, data readiness and a narrow pilot. For an established AI function, review the existing portfolio, operating performance and delivery bottlenecks before adding another experiment. For high-impact or regulated uses, assurance and approval may take priority over launching within 90 days.

This is an applied AI leadership framework. A research-led or model-development Head of AI role also needs a discipline-specific plan for scientific direction, data and model development, compute and technical talent.

Research behind the method

I use external guidance to challenge the plan, not to prescribe a universal timetable. The leadership track reflects organisational accountability, risk management, workforce readiness and ongoing oversight; the dates are my planning structure.

National AI Centre: guidance for AI adoption NIST AI RMF: Govern, Map, Measure and Manage

The principles in practice